Head to head
Clef Omni scores 79.7 and Kev 4B 76.8 on the System One Index. Clef Omni is more accurate on 10 of 16 text datasets, Kev 4B on 5, and 1 are tied. Same cases, same request, same scorer.
| Clef Omni | Kev 4B | |
|---|---|---|
| System One Index | 79.7 | 76.8 |
| English | 78.6 | 78.3 |
| Arabic | 80.5 | 74.8 |
| Vision Index | 72.2 | – |
| Calibration error | 3.4% | 5.1% |
| Median latency | 208 ms | 552 ms |
| Cost per 1,000 calls | $0.074 | $0.010 |
| Clef Omni | Kev 4B | |
|---|---|---|
| AG News English | 89.0 | 90.5 |
| SST-2 English | 97.0 | 92.5 |
| SST-5 English | 51.0 | 52.0 |
| DAIR Emotion English | 56.0 | 54.5 |
| Banking77 English | 96.0 | 88.0 |
| BoolQ English | 90.0 | 91.5 |
| XNLI English English | 82.7 | 90.7 |
| Typed decisions English | 67.0 | 66.3 |
| XNLI French French | 82.7 | 79.3 |
| XNLI Arabic Arabic | 73.3 | 72.7 |
| MASSIVE scenario Arabic | 89.0 | 57.0 |
| MASSIVE intent Arabic | 91.5 | 76.5 |
| OSACT4 offensive Arabic | 90.5 | 88.5 |
| Tweet sentiment Arabic | 50.0 | 63.0 |
| MIRACL rerank Arabic | 83.0 | 79.5 |
| MIRACL pairwise Arabic | 86.5 | 86.5 |
| Clef Omni | Kev 4B | |
|---|---|---|
| SST-2, as an image | 89.5 | – |
| Arabic sentiment, as an image | 28.5 | – |
| Oxford-IIIT Pet | 88.0 | – |
| VQAv2 yes/no | 87.0 | – |
| CountBench | 68.0 | – |
Model pages: Clef Omni · Kev 4B. Latency for API models is the median per call through OpenRouter, timed from one machine.